CultureTalk-ID - Language Steering into Local Languages: leaderboard
CultureTalk-ID (2026) tests cultural commonsense in dialogues grounded in regional Indonesian cultures, in Indonesian and in local languages such as Javanese, Minangkabau and Balinese. This board is the percent of dialogue continuations written in the requested local language when steered from Indonesian, by GlotLID language identification; higher is better.
Metric: GlotLID accuracy (%; share of zero-shot dialogue continuations written in the requested local language when steered from an Indonesian dialogue; higher is better). Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 92.7 | #144 |
| 2 | GPT-5.1 | 73.9 | #229 |
| 3 | Command A | 31.6 | #1274 |
| 4 | Sailor2-8B-Chat | 23.7 | #2396 |
| 5 | gemma2-9B-cpt-sahabatai-v1-instruct | 13.7 | #1488 |
| 6 | Llama 3.1 8B Instruct | 9.2 | #4246 |
| 7 | Gemma 2 9B (IT) | 7.8 | #2262 |
| 8 | Qwen 3 8B | 1.5 | #1830 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=culturetalk-id-language-steering-into-local-languages · How It Works · Data refreshed daily, snapshot 2026-11-13.